Re: KCSAN: data-race in __alloc_file / __alloc_file

From: Eric Dumazet <edumazet@google.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>,
	syzbot <syzbot+3ef049d50587836c0606@syzkaller.appspotmail.com>,
	Marco Elver <elver@google.com>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	syzkaller-bugs <syzkaller-bugs@googlegroups.com>,
	Al Viro <viro@zeniv.linux.org.uk>
Subject: Re: KCSAN: data-race in __alloc_file / __alloc_file
Date: Fri, 8 Nov 2019 09:53:01 -0800	[thread overview]
Message-ID: <CANn89iJJiB6avNtZ1qQNTeJwyjW32Pxk_2CwvEJxgQ==kgY0fA@mail.gmail.com> (raw)
In-Reply-To: <CAHk-=whNBL63qmO176qOQpkY16xvomog5ocvM=9K55hUgAgOPA@mail.gmail.com>

On Fri, Nov 8, 2019 at 9:39 AM Linus Torvalds
<torvalds@linux-foundation.org> wrote:

> I'd hope that there is some way to mark the cases we know about where
> we just have a flag. I'm not sure what KCSAN uses right now - is it
> just the "volatile" that makes KCSAN ignore it, or are there other
> ways to do it?

I dunno, Marco will comment on this.

I personally like WRITE_ONCE() since it adds zero overhead on generated code,
and is the facto accessor we used for many years (before KCSAN was conceived)

>
> "volatile" has huge problems with code generation for gcc. It would
> probably be fine for "not_rcu" in this case, but I'd like to avoid it
> in general otherwise, which is why I wonder if there are other
> options.
>
> But worst comes to worst, I'd be ok with a WRITE_ONCE() and a comment
> about why (and the reason being KCSAN, not the questionable
> optimization).

Ok for a single WRITE_ONCE() with a comment.

Hmm, which questionable optimization are you referring to?